Your AI problem is not a technology problem

Most enterprise AI never reaches the P&L. The pilots work, the demos impress, and the value evaporates somewhere between the proof of concept and the operating model. Last year the proportion of companies abandoning most of their AI initiatives before production more than doubled. The models are not the reason.

BCG's analysis of successful AI transformations puts it plainly: 10 per cent of the work is algorithms, 20 per cent is technology and data, and 70 per cent is people and process. Yet roughly 90 per cent of corporate spend, attention and board reporting chases the technological slice. McKinsey finds that redesigning workflows is the single factor most correlated with bottom-line impact from AI, and that the firms seeing real returns are nearly three times more likely to have fundamentally redesigned how they work. Most reach for the technology first, and most see nothing.

We sell the missing 70 per cent.

The research behind this argument — from METR's capability doubling times to Brynjolfsson's J-curve to six distinct mechanisms of organisational stall — is set out in full in Two Clocks: Why Operating Models, Not Models, Will Decide Who Profits from AI.

What we do

Dromologue is an AI Transformation Incubator. Over three to six months we take your own prioritised use cases and run them as a test bench for redesigning the operating model around AI: who decides, who verifies, who is accountable, where the governance gates sit, who owns the token spend, and how roles change when domain experts can build directly and engineers own the paved road to production.

This is not a pilot factory and not a slide deck. The use cases ship as working capabilities. The wider deliverable is the documented pattern of how your organisation must change to run them, and everything that follows them, at scale.

The work runs across six disciplines: rewriting the policy and procedure estate for a workforce that now includes agents; token cost optimisation as a first-class financial discipline; the culture change that determines whether anything generalises beyond the pilot team; sourcing the new shape of talent; embedded coaching of your senior engineers and architects on real codebases, not training PowerPoints; and skill codification, so that every use case leaves behind a documented, evaluated capability with its own tests, guardrails and policy shape.

What you keep

You keep the working capabilities, the redesigned operating model, the coached engineers, and the codified skills. The skill, not the slide, is the deliverable that compounds. Each one carries its own evaluation harness, so it remains auditable, improvable and re-runnable long after we have gone. Knowledge that would otherwise walk out the door with the consultants stays in your estate, in executable form.

Why now

Three clocks are running. Boards are asking why AI spend is not moving EBIT. The EU AI Act is arriving in stages, with high-risk obligations currently due from August 2026 and its heaviest penalties reaching €35 million or 7 per cent of global turnover, forcing the operating-model conversation regardless of ambition. And as agentic workflows scale, token spend is becoming the fastest-growing cost line in the business, with no operating-model owner. Firms that do the organisational work first take the surplus; the economics of every previous general-purpose technology say so.

Why us

We are operator-executives, not career advisers. We have built enterprise platforms, governance frameworks and engineering cultures at global scale, across three continents, in regulated financial services. We have watched, repeatedly, what happens when firms try to solve this with technology alone. We start in financial services because that is where the operating-model debt is deepest, the regulatory pressure most concrete, and our domain expertise runs strongest, and because in this work, domain knowledge is the variable that bounds the quality of everything AI produces.

The technology is the forcing function. The organisational redesign is actually the answer.